A framework for dynamic load balancing and physical reorganization in parallel database systems
Guardado en:
| Publicado en: | ProQuest Dissertations and Theses (1996) |
|---|---|
| Autor principal: | |
| Publicado: |
ProQuest Dissertations & Theses
|
| Materias: | |
| Acceso en línea: | Citation/Abstract Full Text - PDF |
| Etiquetas: |
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Resumen: | In this dissertation we present a framework for dynamic load balancing in parallel database systems which are based on a shared-nothing architecture and used for on-line query processing where the execution of each query is localized to a single processing node of the system. In such systems, dynamic load balancing can significantly enhance performance if a skewed load distribution develops across the processing nodes. Load distribution in such systems can get skewed either due to a changing user access pattern to the database or due to insertions, deletions and updates of tuples in the database. A skewed distribution of load causes some processing nodes to be over-utilized while other nodes remain underutilized. As a result the processing capacity of the system is diminished to the extent that a single node or a set of processing nodes can become a bottleneck for the entire system. This limitation of parallel database systems must be addressed if such systems are to be effectively deployed in high performance non-stop applications. The proposed framework consists of a new component for the process architecture of parallel database management systems termed the Load Balancing And Reorganization Module (LBARM). The LBARM is composed of four components: (1) the collector component, (2) the analyzer component, (3) the load balancing planning component, and (4) a cost component. The collector component collects load and performance information from the database. If the analyzer detects a degradation in performance of the system it invokes the planner component to prepare a plan for balancing the load of the system using two different types of algorithms. The first type of algorithms perform load balancing by modifying the clients access pattern to the database while the second type perform an on-line physical reorganization of the database. The choice of a particular algorithm for load balancing depends on both the constituent queries of the workload and physical placement of data. The cost component estimates the cost and benefit associated with a load balancing plan using a closed queuing network which models a parallel database system. The load balancing plan is only executed if the cost of its execution is less than the potential benefits to the database system in terms of enhancement in performance. The collector component is also responsible for the execution of a load balancing plan. A simulation to study the behavior of LBARM shows that it can significantly enhance system performance in a parallel database system with dynamically changing user access patterns. |
|---|---|
| ISBN: | 9780591445312 |
| Fuente: | ProQuest Dissertations & Theses Global |